Ugo Lattanzi
Most AI projects die in the demo. I build the ones that reach production.
Twenty years of building systems that stay up when real traffic hits them — and lately of putting generative AI to work inside hospitals, broadcasters, insurers and manufacturers, where the constraints are real and the room for error is small. I sit on the customer's side of the table as often as at the whiteboard.
Work is measured by what moved.
Four things I do on every engagement.
I start on the customer's side of the table
Discovery, scoping, the technical proposal, the architecture defended in front of a board — and the same day, pair-designing with the engineers who have to build it. I can challenge a CTO's assumptions and open the editor in the same meeting.
I design for the organisation that has to adopt it
The most autonomous system is rarely the right one. On a support platform we deliberately left continuous learning out, because the customer needed the knowledge base under their own control. An autonomous system nobody switches on is worth nothing.
If it isn't measured, it isn't finished
Validation programmes run with the domain experts who sign the output, regression suites that assert the arguments a model produces rather than its prose, and feedback captured while the system is working. It keeps the bill down too.
I leave the team able to do it without me
I have taken companies from no AI at all to AI in daily use — my own consultancy first, then a larger one, where a unit of seven now runs inside the process I designed. Enablement isn't a phase at the end of the project. It is the project.
What I actually think about AI.
I am not neutral about this technology. I think it is the largest shift I will see in my working life, and I spent two years proving it on myself before selling it to anyone — rebuilding how my own company wrote code, reviewed it, tested it and documented it, and measuring what actually changed rather than assuming.
The interesting question was never what the model can do. It is what an organisation can absorb.
That is where most of this work actually happens — in trust, in regulation, in who is accountable for the answer, in whether the people on the receiving end will use it on a Tuesday afternoon. The engineering is the easy half.
So I build assistive before autonomous, I keep the human at the last step until the evidence says otherwise, and I measure everything. Not because I am cautious about AI. Because that is how it ends up being used.
Explaining it is half the job.
If this is the problem on your desk, write to me.
There is no CV on this site on purpose — I would rather have the conversation. Ask, and it is in your inbox the same day. I work in Italian, English and German-speaking Switzerland, and I travel.